Content must come from an actual visit, purchase or service experience; the system must not fabricate one.
Trust rule · v1.0
AI can assist expression. It cannot invent a person's experience.
DGP separates authentic experience, AI drafting, user confirmation and publishing into four reviewable steps. Missing any step means the result must not be presented as authentic customer content.
AI may organise text, images or video drafts from the points selected by the user.
The user can review, edit or decline the draft and decide whether to continue.
The user or an explicitly authorised operator performs the publishing action.
Boundaries that must not be crossed
No fabrication or undisclosed publishing
- Do not use prewritten store content to impersonate a customer's experience.
- Do not preselect positive ratings or suppress negative experiences.
- Do not submit reviews or content without the user's knowledge.
- Do not fabricate visits, transactions, views, likes, redemptions or other results.
Benefits must not require a prescribed positive review
- A benefit may support event participation or feedback, but must not require a specified rating or positive wording.
- Review, coupon and membership actions should remain understandable, separate choices.
- Rules, validity periods and restrictions must be shown before the action.
- If a destination platform applies stricter rules, those rules take precedence.
Product controls required before launch
| Control | Minimum requirement | Acceptance evidence |
|---|---|---|
| Notice and permission | Explain the information collected, purpose, retention and publishing destination. | Page capture, permission record and privacy-notice version. |
| Content confirmation | Show the complete draft before publishing and allow editing, regeneration and cancellation. | Flow recording and test cases. |
| Publishing action | Do not bypass the user or platform authorisation to submit content. | Publishing-flow test and platform authorisation record. |
| Benefit design | Do not link a benefit to a specified positive review, like or directed vote. | Campaign rules and staff training record. |
| Correction and complaints | Provide routes for deletion, correction, withdrawal and human review. | Support record and handling target. |
Operator checklist
Before launch
Review platform rules, visitor copy, benefit conditions, privacy information and staff training.
During operation
Sample the drafting and publishing flow, record exceptions and avoid prescribed wording.
After operation
Review complaints, withdrawals, content accuracy and metric definitions, then correct issues promptly.
Reference framework and updates
Reviews and commercial claims
The Chinese operation references the State Administration for Market Regulation's Interim Provisions Against Unfair Competition on the Internet and the Advertising Law of the People's Republic of China. Claims about data, effects and commercial returns require support.
Personal information and facial images
The Chinese operation references the Cyberspace Administration of China's Measures for the Security Management of Facial Recognition Technology Applications and the Personal Information Protection Law of the People's Republic of China. The Chinese source texts are authoritative.